Skip to content

resource_iteration_events table blocks user deletion owing to SET NULL / NOT NULL contradiction

Summary

Customer raised a ticket as they get errors deleting a user. GitLab team members can read more in the tickets (1, 2).

In the PostgreSQL log:

2021-11-30_03:08:21.68660 ERROR: null value in column "user_id" violates not-null constraint\  
2021-11-30_03:08:21.68669 DETAIL: Failing row contains (123, null, 456, null, 789, 2021-02-29 23:59:00.123456+00, 1).\  
2021-11-30_03:08:21.68672 CONTEXT: SQL statement "UPDATE ONLY "public"."resource_iteration_events" SET "user_id" = NULL WHERE $1 OPERATOR(pg_catalog.=) "user_id""\  

In Sidekiq jobs view:

image

Looking at the definition of the table in 13.12.15 (the relevant elements are still the same in %14.6 :

CREATE TABLE resource_iteration_events (
    id bigint NOT NULL,
    user_id bigint NOT NULL,
    issue_id bigint,
    merge_request_id bigint,
    iteration_id bigint,
    created_at timestamp with time zone NOT NULL,
    action smallint NOT NULL
);
ALTER TABLE ONLY resource_iteration_events
    ADD CONSTRAINT fk_rails_501fa15d69 FOREIGN KEY (user_id) REFERENCES users(id) ON DELETE SET NULL;

The user deletion in the users table will attempt to SET NULL the user_id field via the FK constraint, but the field is set not nullable.

Workaround

  1. Run the Rails console

    gitlab-rails console
  2. Reassign these records to the ghost user. In this example, the username (@handle) is fred.delete

    ghostu = User.find_by(username: 'ghost')
    deleteu = User.find_by(username: 'fred.delete')
    ResourceIterationEvent.where(user_id: deleteu.id).update(user_id: ghostu.id)
  3. Delete the user account. With the records assigned to the Ghost user, this will now be successful.

Steps to reproduce

  1. Create a user.
  2. Generate iteration activity using that user, such as assigning an issue to an iteration.
  3. Delete the user.

Example Project

What is the current bug behavior?

User deletion fails if they've generated records in the resource_iteration_events table.

What is the expected correct behavior?

User deletion is successful.

Relevant logs and/or screenshots

gitlabhq_production=# select * from resource_iteration_events;
 id | user_id | issue_id | merge_request_id | iteration_id |          created_at           | action 
----+---------+----------+------------------+--------------+-------------------------------+--------
  1 |       1 |   147206 |                  |            1 | 2021-12-30 09:51:35.598221+00 |      1
  2 |       4 |   147207 |                  |            2 | 2021-12-30 10:18:25.903289+00 |      1

based on rails console cheat sheet

irb(main):008:0> user = User.find_by_id(4)
=> #<User id:4 @zd259193>
irb(main):009:0> users = User.where('id NOT IN (select distinct(user_id) from project_authorizations)')
=> #<ActiveRecord::Relation [#<User id:2 @alert-bot>, #<User id:3 @support-bot>, #<User id:4 @zd259193>]>
irb(main):010:0>  users = User.where('id in (4)')
=> #<ActiveRecord::Relation [#<User id:4 @zd259193>]>
irb(main):011:0> current_user = User.find_by(id: 1)
=> #<User id:1 @root>
irb(main):012:1* users.each do |user|
irb(main):013:1*   DeleteUserWorker.new.perform(current_user.id, user.id)
irb(main):014:0> end
Traceback (most recent call last):
       16: from app/services/users/destroy_service.rb:69:in `execute'
       15: from lib/gitlab/database/query_analyzers/prevent_cross_database_modification.rb:16:in `allow_cross_database_modification_within_transaction'
       14: from lib/gitlab/database/query_analyzers/base.rb:18:in `with_suppressed'
       13: from app/services/users/destroy_service.rb:70:in `block in execute'
       12: from lib/gitlab/database/load_balancing/connection_proxy.rb:77:in `transaction'
       11: from lib/gitlab/database/load_balancing/connection_proxy.rb:125:in `write_using_load_balancer'
       10: from lib/gitlab/database/load_balancing/load_balancer.rb:110:in `read_write'
        9: from lib/gitlab/database/load_balancing/load_balancer.rb:172:in `retry_with_backoff'
        8: from lib/gitlab/database/load_balancing/load_balancer.rb:112:in `block in read_write'
        7: from lib/gitlab/database/load_balancing/connection_proxy.rb:126:in `block in write_using_load_balancer'
        6: from lib/gitlab/database/load_balancing/connection_proxy.rb:67:in `block (2 levels) in <class:ConnectionProxy>'
        5: from lib/gitlab/database/load_balancing/connection_proxy.rb:125:in `write_using_load_balancer'
        4: from lib/gitlab/database/load_balancing/load_balancer.rb:110:in `read_write'
        3: from lib/gitlab/database/load_balancing/load_balancer.rb:172:in `retry_with_backoff'
        2: from lib/gitlab/database/load_balancing/load_balancer.rb:112:in `block in read_write'
        1: from lib/gitlab/database/load_balancing/connection_proxy.rb:126:in `block in write_using_load_balancer'
ActiveRecord::NotNullViolation (PG::NotNullViolation: ERROR:  null value in column "user_id" violates not-null constraint)
DETAIL:  Failing row contains (2, null, 147207, null, 2, 2021-12-30 10:18:25.903289+00, 1).
CONTEXT:  SQL statement "UPDATE ONLY "public"."resource_iteration_events" SET "user_id" = NULL WHERE $1 OPERATOR(pg_catalog.=) "user_id""

Output of checks

Results of GitLab environment info

Expand for output related to GitLab environment info

(For installations with omnibus-gitlab package run and paste the output of:
`sudo gitlab-rake gitlab:env:info`)

(For installations from source run and paste the output of:
`sudo -u git -H bundle exec rake gitlab:env:info RAILS_ENV=production`)

Results of GitLab application Check

Expand for output related to the GitLab application check

(For installations with omnibus-gitlab package run and paste the output of: sudo gitlab-rake gitlab:check SANITIZE=true)

(For installations from source run and paste the output of: sudo -u git -H bundle exec rake gitlab:check RAILS_ENV=production SANITIZE=true)

(we will only investigate if the tests are passing)

Possible fixes

Edited by Ben Prescott_